feat：优化对多模态/非多模态replyer的配置

pytests/A_memorix_test/test_legacy_config_migration.py

@@ -33,3 +33,34 @@ def test_legacy_learning_list_with_numeric_fourth_column_is_migrated():
             "enable_jargon_learning": False,
         },
     ]
+
+
+def test_visual_multimodal_replyer_is_migrated_to_replyer_mode() -> None:
+    payload = {
+        "visual": {
+            "multimodal_replyer": True,
+        }
+    }
+
+    result = try_migrate_legacy_bot_config_dict(payload)
+
+    assert result.migrated is True
+    assert "visual.multimodal_replyer_moved_to_visual.replyer_mode" in result.reason
+    assert result.data["visual"]["replyer_mode"] == "multimodal"
+    assert "multimodal_replyer" not in result.data["visual"]
+
+
+def test_chat_replyer_generator_type_is_migrated_to_replyer_mode() -> None:
+    payload = {
+        "chat": {
+            "replyer_generator_type": "legacy",
+        },
+        "visual": {},
+    }
+
+    result = try_migrate_legacy_bot_config_dict(payload)
+
+    assert result.migrated is True
+    assert "chat.replyer_generator_type_moved_to_visual.replyer_mode" in result.reason
+    assert result.data["visual"]["replyer_mode"] == "text"
+    assert "replyer_generator_type" not in result.data["chat"]

src/chat/replyer/maisaka_generator.py

@@ -1,10 +1,8 @@
-from datetime import datetime
 from typing import Any, Callable, Optional
 
 from src.chat.message_receive.chat_manager import BotChatSession
 from src.common.prompt_i18n import load_prompt
 from src.config.config import global_config
-from src.maisaka.context_messages import SessionBackedMessage
 from src.services.llm_service import LLMServiceClient
 
 from .maisaka_generator_base import BaseMaisakaReplyGenerator
@@ -26,9 +24,6 @@ class MaisakaReplyGenerator(BaseMaisakaReplyGenerator):
             request_type=request_type,
             llm_client_cls=llm_client_cls or LLMServiceClient,
             load_prompt_func=load_prompt_func or load_prompt,
-            enable_visual_message=(
+            enable_visual_message=enable_visual_message,
-                global_config.visual.multimodal_replyer
+            replyer_mode=global_config.visual.replyer_mode,
-                if enable_visual_message is None
-                else enable_visual_message
-            ),
         )

src/chat/replyer/maisaka_generator_base.py

@@ -1,7 +1,7 @@
 import time
 from dataclasses import dataclass, field
 from datetime import datetime
-from typing import Any, Awaitable, Callable, Dict, List, Optional, Tuple
+from typing import Any, Awaitable, Callable, Dict, List, Literal, Optional, Tuple
 
 import random
 
@@ -23,6 +23,7 @@ from src.common.data_models.reply_generation_data_models import (
 from src.common.logger import get_logger
 from src.common.utils.utils_session import SessionUtils
 from src.config.config import global_config
+from src.config.model_configs import ModelInfo
 from src.core.types import ActionInfo
 from src.llm_models.payload_content.message import Message, MessageBuilder, RoleType
 from src.maisaka.context_messages import (
@@ -59,13 +60,15 @@ class BaseMaisakaReplyGenerator:
         request_type: str = "maisaka_replyer",
         llm_client_cls: Any,
         load_prompt_func: Callable[..., str],
-        enable_visual_message: bool,
+        enable_visual_message: Optional[bool],
+        replyer_mode: Literal["text", "multimodal", "auto"],
     ) -> None:
         self.chat_stream = chat_stream
         self.request_type = request_type
         self._llm_client_cls = llm_client_cls
         self._load_prompt = load_prompt_func
         self._enable_visual_message = enable_visual_message
+        self._replyer_mode = replyer_mode
         self.express_model = llm_client_cls(
             task_name="replyer",
             request_type=request_type,
@@ -265,8 +268,9 @@ class BaseMaisakaReplyGenerator:
     def _build_visual_user_message(
         self,
         message: SessionBackedMessage,
+        enable_visual_message: bool,
     ) -> Optional[Message]:
-        if not self._enable_visual_message:
+        if not enable_visual_message:
             return None
 
         raw_message = clone_message_sequence(message.raw_message)
@@ -283,7 +287,11 @@ class BaseMaisakaReplyGenerator:
         )
         return visual_message.to_llm_message()
 
-    def _build_history_messages(self, chat_history: List[LLMContextMessage]) -> List[Message]:
+    def _build_history_messages(
+        self,
+        chat_history: List[LLMContextMessage],
+        enable_visual_message: bool,
+    ) -> List[Message]:
         bot_nickname = global_config.bot.nickname.strip() or "Bot"
         default_user_name = global_config.maisaka.cli_user_name.strip() or "User"
         messages: List[Message] = []
@@ -300,7 +308,7 @@ class BaseMaisakaReplyGenerator:
                     )
                     continue
 
-                visual_message = self._build_visual_user_message(message)
+                visual_message = self._build_visual_user_message(message, enable_visual_message)
                 if visual_message is not None:
                     messages.append(visual_message)
                     continue
@@ -337,6 +345,7 @@ class BaseMaisakaReplyGenerator:
         reply_reason: str,
         expression_habits: str = "",
         stream_id: Optional[str] = None,
+        enable_visual_message: bool = False,
     ) -> List[Message]:
         messages: List[Message] = []
         system_prompt = self._build_system_prompt(
@@ -348,10 +357,21 @@ class BaseMaisakaReplyGenerator:
         instruction = self._build_reply_instruction()
 
         messages.append(MessageBuilder().set_role(RoleType.System).add_text_content(system_prompt).build())
-        messages.extend(self._build_history_messages(chat_history))
+        messages.extend(self._build_history_messages(chat_history, enable_visual_message))
         messages.append(MessageBuilder().set_role(RoleType.User).add_text_content(instruction).build())
         return messages
 
+    def _resolve_enable_visual_message(self, model_info: Optional[ModelInfo] = None) -> bool:
+        if self._enable_visual_message is not None:
+            return self._enable_visual_message
+        if self._replyer_mode == "multimodal":
+            if model_info is not None and not model_info.visual:
+                raise ValueError(f"replyer_mode=multimodal，但模型 '{model_info.name}' 未开启 visual，无法使用多模态 replyer")
+            return True
+        if self._replyer_mode == "text":
+            return False
+        return bool(model_info.visual) if model_info is not None else False
+
     def _resolve_session_id(self, stream_id: Optional[str]) -> str:
         if stream_id:
             return stream_id
@@ -494,7 +514,19 @@ class BaseMaisakaReplyGenerator:
         show_replyer_prompt = bool(getattr(global_config.debug, "show_replyer_prompt", False))
         show_replyer_reasoning = bool(getattr(global_config.debug, "show_replyer_reasoning", False))
 
-        def message_factory(_client: object) -> List[Message]:
+        def message_factory(_client: object, model_info: Optional[ModelInfo] = None) -> List[Message]:
+            nonlocal prompt_ms, prompt_preview, request_messages
+            prompt_started_at = time.perf_counter()
+            request_messages = self._build_request_messages(
+                chat_history=filtered_history,
+                reply_message=reply_message,
+                reply_reason=reply_reason or "",
+                expression_habits=merged_expression_habits,
+                stream_id=stream_id,
+                enable_visual_message=self._resolve_enable_visual_message(model_info),
+            )
+            prompt_ms = round((time.perf_counter() - prompt_started_at) * 1000, 2)
+            prompt_preview = PromptCLIVisualizer._build_prompt_dump_text(request_messages)
             return request_messages
 
         result.completion.request_prompt = prompt_preview
@@ -531,6 +563,8 @@ class BaseMaisakaReplyGenerator:
             )
             return finalize(False)
 
+        result.completion.request_prompt = prompt_preview
+        result.request_messages = serialize_prompt_messages(request_messages)
         llm_ms = round((time.perf_counter() - llm_started_at) * 1000, 2)
         response_text = (generation_result.response or "").strip()
         result.success = bool(response_text)

src/chat/replyer/replyer_manager.py

@@ -1,4 +1,4 @@
-from typing import TYPE_CHECKING, Any, Dict, Optional
+from typing import Any, Dict, Optional
 
 from src.chat.message_receive.chat_manager import BotChatSession, chat_manager as _chat_manager
 from src.config.config import global_config
@@ -6,10 +6,6 @@ from src.common.logger import get_logger
 
 from .maisaka_generator import MaisakaReplyGenerator
 
-if TYPE_CHECKING:
-    from src.chat.replyer.group_generator import DefaultReplyer
-    from src.chat.replyer.private_generator import PrivateReplyer
-
 logger = get_logger("ReplyerManager")
 
 
@@ -22,7 +18,7 @@ class ReplyerManager:
     @staticmethod
     def _get_maisaka_generator_type() -> str:
         """返回当前配置下 Maisaka replyer 的消息模式。"""
-        return "multimodal" if global_config.visual.multimodal_replyer else "legacy"
+        return global_config.visual.replyer_mode
 
     def get_replyer(
         self,
@@ -30,7 +26,7 @@ class ReplyerManager:
         chat_id: Optional[str] = None,
         request_type: str = "replyer",
         replyer_type: str = "default",
-    ) -> Optional["DefaultReplyer | PrivateReplyer | Any"]:
+    ) -> Optional[MaisakaReplyGenerator]:
         """按会话和 replyer 类型获取实例。"""
         stream_id = chat_stream.session_id if chat_stream else chat_id
         if not stream_id:

src/common/data_models/llm_service_data_models.py

@@ -14,7 +14,6 @@ from src.llm_models.payload_content.resp_format import RespFormat
 from src.llm_models.payload_content.tool_option import ToolCall, ToolDefinitionInput
 
 if TYPE_CHECKING:
-    from src.llm_models.model_client.base_client import BaseClient
     from src.llm_models.payload_content.message import Message
 
 
@@ -24,7 +23,7 @@ PromptMessage: TypeAlias = Dict[str, Any]
 PromptInput: TypeAlias = str | List[PromptMessage]
 """统一的提示输入类型。"""
 
-MessageFactory: TypeAlias = Callable[["BaseClient"], List["Message"]]
+MessageFactory: TypeAlias = Callable[..., List["Message"]]
 """统一的消息工厂类型。"""
 
 

src/config/config.py

@@ -54,7 +54,7 @@ CONFIG_DIR: Path = PROJECT_ROOT / "config"
 BOT_CONFIG_PATH: Path = (CONFIG_DIR / "bot_config.toml").resolve().absolute()
 MODEL_CONFIG_PATH: Path = (CONFIG_DIR / "model_config.toml").resolve().absolute()
 MMC_VERSION: str = "1.0.0"
-CONFIG_VERSION: str = "8.5.5"
+CONFIG_VERSION: str = "8.6.0"
 MODEL_CONFIG_VERSION: str = "1.14.0"
 
 logger = get_logger("config")

src/config/legacy_migration.py

@@ -253,6 +253,8 @@ def _migrate_target_item_list(parent: dict[str, Any], key: str) -> bool:
     raw = _as_list(parent.get(key))
     if raw is None:
         return False
+    if not raw:
+        return False
     if raw and all(isinstance(i, dict) for i in raw):
         return False
     targets: list[dict[str, str]] = []
@@ -285,18 +287,18 @@ def _migrate_extra_prompt_list(exp: dict[str, Any], key: str) -> bool:
     return True
 
 
-def _parse_multimodal_replyer(v: Any) -> Optional[bool]:
+def _parse_replyer_mode(v: Any) -> Optional[str]:
     """兼容旧 replyer_generator_type 到布尔开关的迁移。"""
     if isinstance(v, bool):
-        return v
+        return "multimodal" if v else "text"
     if not isinstance(v, str):
         return None
 
     normalized_value = v.strip().lower()
-    if normalized_value == "multimodal":
+    if normalized_value in {"text", "multimodal", "auto"}:
-        return True
+        return normalized_value
     if normalized_value == "legacy":
-        return False
+        return "text"
     return None
 
 
@@ -403,14 +405,23 @@ def try_migrate_legacy_bot_config_dict(data: dict[str, Any]) -> MigrationResult:
             migrated_any = True
             reasons.append("chat.multimodal_planner_moved_to_visual.multimodal_planner")
 
+    if visual is not None and "multimodal_replyer" in visual:
+        replyer_mode = _parse_replyer_mode(visual.get("multimodal_replyer"))
+        if "replyer_mode" not in visual and replyer_mode is not None:
+            visual["replyer_mode"] = replyer_mode
+        if "replyer_mode" in visual:
+            visual.pop("multimodal_replyer", None)
+            migrated_any = True
+            reasons.append("visual.multimodal_replyer_moved_to_visual.replyer_mode")
+
     if visual is not None and "replyer_generator_type" in chat:
-        multimodal_replyer = _parse_multimodal_replyer(chat["replyer_generator_type"])
+        replyer_mode = _parse_replyer_mode(chat["replyer_generator_type"])
-        if "multimodal_replyer" not in visual and multimodal_replyer is not None:
+        if "replyer_mode" not in visual and replyer_mode is not None:
-            visual["multimodal_replyer"] = multimodal_replyer
+            visual["replyer_mode"] = replyer_mode
-        if "multimodal_replyer" in visual:
+        if "replyer_mode" in visual:
             chat.pop("replyer_generator_type", None)
             migrated_any = True
-            reasons.append("chat.replyer_generator_type_moved_to_visual.multimodal_replyer")
+            reasons.append("chat.replyer_generator_type_moved_to_visual.replyer_mode")
 
     maisaka = _as_dict(data.get("maisaka"))
     mem = _as_dict(data.get("memory"))

src/config/official_configs.py

@@ -152,16 +152,16 @@ class VisualConfig(ConfigBase):
             "x-icon": "image",
         },
     )
-    """是否直接输入图片"""
+    """是否启用多模态planner"""
 
-    multimodal_replyer: bool = Field(
+    replyer_mode: Literal["text", "multimodal", "auto"] = Field(
-        default=False,
+        default="auto",
         json_schema_extra={
-            "x-widget": "switch",
+            "x-widget": "select",
             "x-icon": "git-branch",
         },
     )
-    """是否启用 Maisaka 多模态 replyer 生成器"""
+    """回复器模式，auto根据模型信息自动选择，text为纯文本模式，multimodal为多模态模式"""
 
     visual_style: str = Field(
         default="请用中文描述这张图片的内容。如果有文字，请把文字描述概括出来，请留意其主题，直观感受，输出为一段平文本，最多30字，请注意不要分点，就输出一段文本",

src/llm_models/utils_model.py

@@ -3,6 +3,7 @@ from enum import Enum
 from typing import Any, Callable, Dict, List, Optional, Set, Tuple
 
 import asyncio
+import inspect
 import random
 import re
 import time
@@ -910,6 +911,10 @@ class LLMOrchestrator:
             model_info, api_provider, client = self._select_model(exclude_models=failed_models_this_request)
             message_list = []
             if message_factory:
+                parameter_count = len(inspect.signature(message_factory).parameters)
+                if parameter_count >= 2:
+                    message_list = message_factory(client, model_info)
+                else:
                     message_list = message_factory(client)
             try:
                 request = self._build_client_request(

src/maisaka/builtin_tool/send_emoji.py

@@ -2,11 +2,11 @@
 
 from datetime import datetime
 from io import BytesIO
-import math
 from random import sample
 from typing import Any, Dict, Optional
 
 import asyncio
+import math
 
 from PIL import Image as PILImage
 from PIL import ImageDraw, ImageFont
@@ -20,12 +20,14 @@ from src.common.logger import get_logger
 from src.config.config import global_config
 from src.core.tooling import ToolExecutionContext, ToolExecutionResult, ToolInvocation, ToolSpec
 from src.llm_models.payload_content.resp_format import RespFormat, RespFormatType
+from src.llm_models.payload_content.message import MessageBuilder, RoleType
 from src.maisaka.context_messages import (
     LLMContextMessage,
     ReferenceMessage,
     ReferenceMessageType,
     SessionBackedMessage,
 )
+from src.plugin_runtime.hook_payloads import serialize_prompt_messages
 
 from .context import BuiltinToolRuntimeContext
 
@@ -270,6 +272,7 @@ def _build_send_emoji_prompt_preview(
 def _build_send_emoji_monitor_detail(
     *,
     prompt_text: str = "",
+    request_messages: Optional[list[dict[str, Any]]] = None,
     reasoning_text: str = "",
     output_text: str = "",
     metrics: Optional[Dict[str, Any]] = None,
@@ -280,6 +283,8 @@ def _build_send_emoji_monitor_detail(
     detail: Dict[str, Any] = {}
     if prompt_text.strip():
         detail["prompt_text"] = prompt_text.strip()
+    if isinstance(request_messages, list) and request_messages:
+        detail["request_messages"] = request_messages
     if reasoning_text.strip():
         detail["reasoning_text"] = reasoning_text.strip()
     if output_text.strip():
@@ -394,6 +399,16 @@ async def _select_emoji_with_sub_agent(
         grid_columns=grid_columns,
         sampled_emojis=sampled_emojis,
     )
+    request_messages = [
+        MessageBuilder().set_role(RoleType.System).add_text_content(system_prompt).build(),
+    ]
+    prompt_llm_message = prompt_message.to_llm_message()
+    if prompt_llm_message is not None:
+        request_messages.append(prompt_llm_message)
+    candidate_llm_message = candidate_message.to_llm_message()
+    if candidate_llm_message is not None:
+        request_messages.append(candidate_llm_message)
+    serialized_request_messages = serialize_prompt_messages(request_messages)
 
     selection_started_at = datetime.now()
     response = await tool_ctx.runtime.run_sub_agent(
@@ -422,6 +437,7 @@ async def _select_emoji_with_sub_agent(
         if selection_metadata is not None:
             selection_metadata["monitor_detail"] = _build_send_emoji_monitor_detail(
                 prompt_text=prompt_preview,
+                request_messages=serialized_request_messages,
                 output_text=response.content or "",
                 metrics=selection_metrics,
                 extra_sections=[{
@@ -436,6 +452,7 @@ async def _select_emoji_with_sub_agent(
         selection_metadata["reason"] = selection.reason.strip()
         selection_metadata["monitor_detail"] = _build_send_emoji_monitor_detail(
             prompt_text=prompt_preview,
+            request_messages=serialized_request_messages,
             reasoning_text=selection.reason,
             output_text=response.content or "",
             metrics=selection_metrics,

src/maisaka/display/prompt_cli_renderer.py

@@ -440,6 +440,9 @@ class PromptCLIVisualizer:
         return (
             "<div class='image-card'>"
             f"<div class='image-meta'>图片 image/{html.escape(normalized_format)} {html.escape(size_text)}</div>"
+            f"<a class='image-preview-link' href='{html.escape(file_uri, quote=True)}'>"
+            f"<img class='image-preview' src='{html.escape(file_uri, quote=True)}' alt='图片预览' />"
+            "</a>"
             f"<div class='image-path'>{html.escape(str(file_path))}</div>"
             f"<a class='image-link' href='{html.escape(file_uri, quote=True)}'>打开图片</a>"
             "</div>"
@@ -727,6 +730,22 @@ class PromptCLIVisualizer:
       font-family: "Cascadia Mono", "JetBrains Mono", "Consolas", monospace;
       word-break: break-all;
     }}
+    .image-preview-link {{
+      display: block;
+      margin-top: 10px;
+    }}
+    .image-preview {{
+      display: block;
+      max-width: min(100%, 560px);
+      max-height: 420px;
+      width: auto;
+      height: auto;
+      border-radius: 12px;
+      border: 1px solid #dbe4f0;
+      background: #fff;
+      box-shadow: 0 8px 20px rgba(15, 23, 42, 0.08);
+      object-fit: contain;
+    }}
     .image-link {{
       display: inline-block;
       margin-top: 8px;

src/services/generator_service.py

@@ -11,8 +11,7 @@ from typing import Any, Dict, List, Optional, Tuple, TYPE_CHECKING
 from rich.traceback import install
 
 from src.chat.message_receive.chat_manager import BotChatSession
-from src.chat.replyer.group_generator import DefaultReplyer
+from src.chat.replyer.maisaka_generator import MaisakaReplyGenerator
-from src.chat.replyer.private_generator import PrivateReplyer
 from src.chat.replyer.replyer_manager import replyer_manager
 from src.chat.utils.utils import process_llm_response
 from src.common.data_models.message_component_data_model import MessageSequence, TextComponent
@@ -38,7 +37,7 @@ def _get_replyer(
     chat_stream: Optional[BotChatSession] = None,
     chat_id: Optional[str] = None,
     request_type: str = "replyer",
-) -> Optional[DefaultReplyer | PrivateReplyer]:
+) -> Optional[MaisakaReplyGenerator]:
     """获取回复器对象"""
     if not chat_id and not chat_stream:
         raise ValueError("chat_stream 和 chat_id 不可均为空")